llm-gemini-youtube

LLM plugin for agentic YouTube video understanding with Gemini.

The plugin uses the Gemini Interactions API and enables agentic processing for every video. Gemini can selectively inspect the frames, audio, and transcript needed to answer the prompt instead of processing the whole video at a fixed frame rate.

Project history

• The 0.0.x series provided server-side YouTube URL processing. That capability was later adopted upstream by simonw/llm-gemini in PR #112: Add support for media_resolution and processing Youtube URLs server-side.
• Starting with the 0.1.x series, this project was rebooted to focus exclusively on Google's agentic video understanding with Gemini.

Installation

Install this plugin in the same environment as LLM.

llm install llm-gemini-youtube

Usage

llm -m gemini-3.7-flash-yt -a 'https://www.youtube.com/watch?v=9hE5-98ZeCg' 'Can you summarize this video?'

llm -m gemini-3.6-flash-yt -a 'https://www.youtube.com/watch?v=9hE5-98ZeCg' 'What are the examples given at 01:05 and 01:19 supposed to show us?'

llm -m gemini-3.5-flash-lite-yt -a 'https://www.youtube.com/watch?v=9hE5-98ZeCg' 'Transcribe the audio from this video, giving timestamps for salient events in the video. Also provide visual descriptions.'

Supported models are gemini-3.7-flash-yt, gemini-3.6-flash-yt, and gemini-3.5-flash-lite-yt. Configure the API key using LLM's standard key mechanism:

llm keys set gemini

Google Cloud

The Google Gen AI SDK can route requests through Gemini Enterprise Agent Platform using its standard environment variables:

export GOOGLE_GENAI_USE_ENTERPRISE=1
export GOOGLE_CLOUD_PROJECT='your-project-id'
export GOOGLE_CLOUD_LOCATION='us-central1'
gcloud auth application-default login

The Gemini key must still be configured for LLM. When the Enterprise variables include a project and location, the SDK selects Google Cloud credentials instead of the API key.

Development

To set up this plugin locally, first checkout the code. Then create a new virtual environment:

cd llm-gemini-youtube
python -m venv venv
source venv/bin/activate

Now install the dependencies and test dependencies:

python -m pip install -e '.[test]'

To run the tests:

python -m pytest